A founder-led, sponsor-backed software company with a monthly board and a first institutional CFO. We built a reporting agent on a semantic layer over the ERP and planning data — it writes the variance narrative, assembles the pack on demand, and answers scenario questions live, in the room.
The CFO and two analysts spent the better part of a week each period assembling a static, backward-looking deck by hand from exported spreadsheets. By the time the pack was bound, the data was three weeks stale. The board's what-if questions couldn't be answered in the room — they became homework. Any sponsor follow-up meant a day of rework.
The Pack, On Demand
ARR · NRR · Gross margin
Cash · Bookings · Pipeline
vs. budget · vs. prior year
Drafted in the house style
Retention by vintage
Expansion vs. churn
Driver-based models
Price · volume · churn · FX
The standard pack builds itself from the same layer, every time — KPI summary, variance narrative, cohort view, scenario appendix. The CFO edits the narrative; the CFO doesn't draft it. Every number in the pack traces to its source.
What We Built
Against budget and against prior year, in the house style the board already knows. The agent reads the actuals as they land, names the drivers, and drafts the story — not a table of deltas, a paragraph a director can read on a plane.
The CFO’s job moved from assembling to editing. What used to consume the front half of every month now takes a red pen and an hour.
A carve-out inherits someone else’s definitions. We built a single semantic layer over the ERP and planning data — one agreed meaning for ARR, churn, and margin — so the agent and the humans argue about the business, never about whose number is right.
Every figure in the pack traces back through that layer to its source. When a director asks “where does this come from?”, the answer is one click, not one week.
This is the part that changed the meeting. Driver-based models sit behind the pack — price, volume, churn, FX — and the agent answers what-if questions in plain language, with the drivers it moved shown alongside the answer.
Board questions stopped becoming homework. Sponsor follow-ups stopped costing a day. The conversation moved from what happened to what to do next.
The Narrative, With Receipts
“EBITDA finished $0.5M below budget. Volume and pricing outperformed plan by $3.0M combined, more than offset by elevated churn in the mid-market cohort ($2.6M) and currency headwinds on European contracts ($0.9M). Churn is concentrated in two renewal classes — detail on page 9.”
The Part That Changed The Meeting
A director asks. The agent rebuilds the driver model and answers — with the number, the chart, and the drivers it moved. The CFO interrogates the model live instead of promising a follow-up.
What does EBITDA look like if we push price 3% in Q3?
We didn't make the pack faster to build. We made it answer back.
The Impact
Board-pack preparation went from five days to half a day, with the pack ready on day six after close. The two analysts who lived inside roll-forwards now do actual FP&A — pricing analysis, cohort economics, the work the sponsor actually wants.
The board stopped reviewing three-week-old numbers. What-if questions get answered while the directors are still in their seats, from the same models that produced the pack — so the answer and the report can never disagree.
For a first institutional CFO in front of a monthly board, that changed the job: less time defending the numbers, more time directing the company.
We design and deploy AI-native systems for companies moving fast in competitive markets.
Talk to LightCI